see __init__.py


TODO: - add an sgd with annealed learning rate (the annealing schedule is implemented via
shared variables and an update schedule

Wishlist:
    - sgd 
    - momentum
    - annealing schedule
    - conjugate methods
    - l-bfgs
    - hessian free
    - tonga
    - smth by nick schraudolff (sp?)

Theano needs lazy if for these to be efficient


Early stopping heuristics as update expressions too?